Regression point annotation can be imported from an Excel file.


Two scenarios are supported:

A) Each of our scans contains only one object

B) There are multiple objects per scan


The scenario A) is the most straightforward and recommeded as a best practice when organizing your experiments.The main advantage is that, having a single object per image, we have more freedom in splitting our data set in training and test sets or performing cross-validation.


The scenario B) is described in detail in this section.


To import point annotation, we need:

1. An existing classifier (Step 1) and object segmentation setup (Step 2).

2. Excel file with scan names and corresponding annotation values


In this example, we use a poweder data set. It contains plastic vials with a mixture of two powders. Each scan contains a single vial.


We wish to import mixing percentages from an Excel file displayed below. It contains image name in the column and the powder concentration percenteage.



We can open Import dialog using the Regression / Import point meta-data menu



Import dialog

The import dialog has two parts. The left side defines the meta-data information from our Excel file. The right side maps the images in our workspace to anotation values in the Excel file.



Loading annotation from Excel file



With a button , we load the desired Excel file. Then, we specify the first cell of the Excel file column with the scan names. In our Excel screenshot above, you can see that the scan names start at B7. Then, for each of our regression variables, we specify the first cell where the values are located. In our situation, this will be L7. Finally, we click the Refresh button to load the content into the list box below.


Our import dialog will now look like this:


TIP Note, that the concentration percentages in our example are part of the scan names. This is the best practice when regression meta-data is known at the time of scanning. In these situations, we can generate our Excel file by exporting object information from perClas Mira and adding an extra column extracting the regression values from the filenames.